Working with geospatial data

Geospatial analytics let you analyze geographic data in BigQuery. Geographic data is also known as geospatial data.

Common types of objects when working with geospatial data include the following:

  • A geometry represents a surface area on the Earth. It is often described using points, lines, polygons, or a collection of points, lines, and polygons. A geometry collection is a geometry that represents the spatial union of all shapes in the collection.
  • A spatial feature represents a logical spatial object. It combines a geometry with additional attributes that are application-specific.
  • A spatial feature collection is a set of spatial features.

In BigQuery, the GEOGRAPHY data type represents a geometry value or geometry collection. To represent spatial features, create a table with a GEOGRAPHY column for the geometry plus additional columns for the attributes. Each row of the table is a spatial feature, and the entire table represents a spatial feature collection.

The GEOGRAPHY data type describes a point set on the Earth's surface. A point set is a set of points, lines, and polygons on the WGS84 reference spheroid, with geodesic edges. You can use the GEOGRAPHY data type by calling one of the GoogleSQL geography functions.

Loading geospatial data

Single points on Earth can be described by just a longitude, latitude pair. For example, you can load a CSV file that contains longitude and latitude values and then use the ST_GEOGPOINT function to convert them into GEOGRAPHY values.

For more complex geographies, you can load the following geospatial data formats into a GEOGRAPHY column:

  • Well-known text (WKT)
  • Well-known binary (WKB)
  • GeoJSON
  • GeoParquet

Loading WKT or WKB data

WKT is a text format for describing individual geometry shapes using points, lines, polygons with optional holes, or a collection of points, lines, or polygons. WKB is the binary version of the WKT format. WKB can be hex encoded for formats that don't support binary data, like JSON.

For example, the following defines a point in WKT:

POINT(-121 41)

To describe a spatial feature, WKT is usually embedded in a container file format, such as a CSV file, or in a database table. A file row or a table row usually corresponds to the spatial feature. The whole file or the whole table corresponds to the feature collection. To load WKT data into BigQuery, provide a schema that specifies a GEOGRAPHY column for the geospatial data.

For example, you might have a CSV file that contains the following data:

"POLYGON((-124.49 47.35,-124.49 40.73,-116.49 40.73,-116.49 47.35,-124.49 47.35))",poly1
"POLYGON((-85.6 31.66,-85.6 24.29,-78.22 24.29,-78.22 31.66,-85.6 31.66))",poly2
"POINT(1 2)",point1

You can load this file by running the bq command-line tool load command:

bq load --source_format=CSV \
  --schema="geography:GEOGRAPHY,name:STRING" \
  mydataset.mytable filename1.csv

For more information about loading data in BigQuery, see Introduction to loading data.

To stream WKT data to an existing BigQuery table with a GEOGRAPHY column, serialize the data as a string in the API request.

bq

Run the bq command-line tool insert command:

echo '{"geo": "LINESTRING (-118.4085 33.9416, -73.7781 40.6413)"}' \
    | bq insert my_dataset.geo_table

Python

Before trying this sample, follow the Python setup instructions in the BigQuery quickstart using client libraries. For more information, see the BigQuery Python API reference documentation.

To authenticate to BigQuery, set up Application Default Credentials. For more information, see Set up authentication for client libraries.

from google.cloud import bigquery
import shapely.geometry
import shapely.wkt

bigquery_client = bigquery.Client()

# This example uses a table containing a column named "geo" with the
# GEOGRAPHY data type.
table_id = "my-project.my_dataset.my_table"

# Use the Shapely library to generate WKT of a line from LAX to
# JFK airports. Alternatively, you may define WKT data directly.
my_geography = shapely.geometry.LineString(
    [(-118.4085, 33.9416), (-73.7781, 40.6413)]
)
rows = [
    # Convert data into a WKT string.
    {"geo": shapely.wkt.dumps(my_geography)},
]

#  table already exists and has a column
# named "geo" with data type GEOGRAPHY.
errors = bigquery_client.insert_rows_json(table_id, rows)
if errors:
    raise RuntimeError(f"row insert failed: {errors}")
else:
    print(f"wrote 1 row to {table_id}")

For more information about streaming data in BigQuery, see Streaming data into BigQuery.

You can also convert a WKT text string into a GEOGRAPHY value by using the ST_GeogFromText function.

Loading GeoJSON data

GeoJSON is a JSON-based format for geometries and spatial features. For example, the following defines a point in GeoJSON:

{ "type": "Point", "coordinates": [-121,41] }

GeoJSON data can contain any of the following object types:

  • Geometry objects. A geometry object is a spatial shape, described as a union of points, lines, and polygons with optional holes.
  • Feature objects. A feature object contains a geometry plus additional name/value pairs, whose meaning is application-specific.
  • Feature collections. A feature collection is a set of feature objects.

There are two ways to load GeoJSON data into BigQuery:

Loading newline-delimited GeoJSON files

A newline-delimited GeoJSON file contains a list of GeoJSON feature objects, one per line in the file. A GeoJSON feature object is a JSON object with the following members:

  • type. For feature objects, the value must be Feature. BigQuery validates the value but does not include it in the table schema.

  • geometry. The value is a GeoJSON Geometry object or null. BigQuery converts this member into a GEOGRAPHY value.

  • properties. The value is any JSON object or null. If the value isn't null, then BigQuery loads each member of the JSON object as a separate table column. For more information about how BigQuery parses JSON data types, see Details of loading JSON data.

  • id. Optional. If present, the value is either a string or a number. BigQuery loads this value into a column named id.

If the feature object contains other members that are not listed here, then BigQuery converts those members directly into table columns.

You can load a newline-delimited GeoJSON file by using the bq command-line tool's bq load command, as follows:

bq load \
 --source_format=NEWLINE_DELIMITED_JSON \
 --json_extension=GEOJSON \
 --autodetect \
 DATASET.TABLE \
 FILE_PATH_OR_URI

Replace the following:

  • DATASET is the name of your dataset.
  • TABLE is the name of the destination table.
  • FILE_PATH_OR_URI is a path to a local file or a Cloud Storage URI.

The previous example enables schema auto-detection. For more control over how BigQuery converts the values inside the properties object, you can provide an explicit schema instead. For more information, see Specify schemas. If you provide an explicit schema, then don't include a top-level type column in the schema definition. For each member of the properties member, define separate columns, not a single nested column.

As defined by RFC 7946, a complete GeoJSON data structure is a single JSON object. Many systems export GeoJSON data as a single FeatureCollection object that contains all of the geometries. To load this format into BigQuery, you must convert the file by removing the root-level FeatureCollection object and splitting the individual feature objects into separate lines. For example, the following command uses the jq command-line tool to split a GeoJSON file into newline- delimited format:

cat ~/file1.json | jq -c '.features[]' > converted.json

Creating an external table from a newline-delimited GeoJSON file

You can query a newline-delimited GeoJSON file stored in Cloud Storage by creating an external table. To create the external table, use the CREATE EXTERNAL TABLE DDL statement. In the OPTIONS clause, set the formatoption to NEWLINE_DELIMITED_JSON and the json_extension option to GEOJSON.

Example:

CREATE EXTERNAL TABLE mydataset.table1 OPTIONS (
  format="NEWLINE_DELIMITED_JSON",
  json_extension = 'GEOJSON',
  uris = ['gs://mybucket/geofile.json']
);

Loading GeoJSON geometry data

Geospatial analytics supports loading individual GeoJSON geometry objects that are embedded as text strings in other file types. For example, you can load a CSV file where one of the columns contains a GeoJSON geometry object.

To load this type of GeoJSON data into BigQuery, provide a schema that specifies a GEOGRAPHY column for the GeoJSON data. You must manually provide the schema. Otherwise, if auto-detection is enabled, then BigQuery loads the data as a STRING value.

Geospatial analytics does not support loading GeoJSON feature objects or feature collections using this approach. If you need to load feature objects, then consider using newline-delimited GeoJSON files.

To stream GeoJSON data to an existing BigQuery table with a GEOGRAPHY column, serialize the data as a string in the API request.

bq

Run the bq command-line tool insert command:

echo '{"geo": "{\"type\": \"LineString\", \"coordinates\": [[-118.4085, 33.9416], [-73.7781, 40.6413]]}"}' \
  | bq insert my_dataset.geo_table

Python

Before trying this sample, follow the Python setup instructions in the BigQuery quickstart using client libraries. For more information, see the BigQuery Python API reference documentation.

To authenticate to BigQuery, set up Application Default Credentials. For more information, see Set up authentication for client libraries.

import geojson
from google.cloud import bigquery

bigquery_client = bigquery.Client()

# This example uses a table containing a column named "geo" with the
# GEOGRAPHY data type.
table_id = "my-project.my_dataset.my_table"

# Use the python-geojson library to generate GeoJSON of a line from LAX to
# JFK airports. Alternatively, you may define GeoJSON data directly, but it
# must be converted to a string before loading it into BigQuery.
my_geography = geojson.LineString([(-118.4085, 33.9416), (-73.7781, 40.6413)])
rows = [
    # Convert GeoJSON data into a string.
    {"geo": geojson.dumps(my_geography)}
]

#  table already exists and has a column
# named "geo" with data type GEOGRAPHY.
errors = bigquery_client.insert_rows_json(table_id, rows)
if errors:
    raise RuntimeError(f"row insert failed: {errors}")
else:
    print(f"wrote 1 row to {table_id}")

You can also convert a GeoJSON geometry object into a GEOGRAPHY value by using the ST_GEOGFROMGEOJSON function. For example, you can store the geometries as STRING values and then run a query that calls ST_GEOGFROMGEOJSON.

Loading GeoParquet files

GeoParquet is a specification that adds geospatial types to the Parquet file format. GeoParquet includes metadata that provides definite semantics to the contained geospatial data, avoiding the interpretation issues that occur with other geospatial data formats.

When loading Parquet files, BigQuery checks for GeoParquet metadata. If GeoParquet metadata exists, BigQuery loads all of the columns it describes into a corresponding GEOGRAPHY column by default. For more information about loading Parquet files, see Loading Parquet data.

Creating an external table from GeoParquet data

External tables that reference GeoParquet files map relevant columns to the GEOGRAPHY type.

Statistics available in the GeoParquet file (bbox, covering) aren't used to accelerate queries over an external table.

Coordinate systems and edges

In geospatial analytics, points are positions on the surface of a WGS84 spheroid, expressed as longitude and geodetic latitude. An edge is a spherical geodesic between two endpoints. (That is, edges are the shortest path on the surface of a sphere.)

The WKT format does not provide a coordinate system. When loading WKT data, geospatial analytics assumes the data uses WGS84 coordinates with spherical edges. Make sure your source data matches that coordinate system, unless the geographies are small enough that the difference between spherical and planar edges can be ignored.

GeoJSON explicitly uses WGS84 coordinates with planar edges. When loading GeoJSON data, geospatial analytics converts planar edges to spherical edges. Geospatial analytics adds additional points to the line as necessary, so that the converted sequence of edges remains within 10 meters of the original line. This process is known as tessellation or non-uniform densification. You cannot directly control the tessellation process.

To load geographies with spherical edges, use WKT. To load geographies with planar edges, often called geometries, it's simplest to use GeoJSON. However, if your geometry data is already in WKT format, another option is to load the data as a STRING type and then use the ST_GEOGFROMTEXT function to convert to GEOGRAPHY values. Set the planar parameter to TRUE to interpret the data as planar.

GeoParquet files include metadata about the coordinate system and edges that were used to create the data. When reading GeoParquet files with planar edges, geospatial analytics converts planar edges to spherical edges. GeoParquet files with coordinate systems other than WGS84 are rejected.

When choosing an interchange format, be sure to understand the coordinate system used by your source data. Most systems either explicitly support parsing geography (as opposed to geometry) from WKT, or else they assume planar edges.

Your coordinates should be longitude first, latitude second. If the geography has any long segments or edges then they must be tessellated, because geospatial analytics interprets them as spherical geodesics, which may not correspond to the coordinate system where your data originated.

Polygon orientation

On a sphere, every polygon has a complementary polygon. For example, a polygon that describes the Earth's continents would have a complementary polygon that describes the Earth's oceans. Because the two polygons are described by the same boundary rings, rules are required to resolve the ambiguity around which of the two polygons is described by a given WKT string.

When you load WKT and WKB strings from files or by using streaming ingestion, geospatial analytics assumes the polygons in the input are oriented as follows: If you traverse the boundary of the polygon in the order of the input vertices, the interior of the polygon is on the left. Geospatial analytics uses the same rule when exporting geography objects to WKT and WKB strings.

If you use the ST_GeogFromText function to convert a WKT string to a GEOGRAPHY value, the oriented parameter specifies how the function determines the polygon:

  • FALSE: Interpret the input as the polygon with the smaller area. This is the default behavior.

  • TRUE: Use the left-hand orientation rule described previously. This option allows you to load polygons with an area larger than a hemisphere.

Because GeoJSON strings are defined on a planar map, the orientation can be determined without ambiguity, even if the input does not follow the orientation rule defined in the GeoJSON format specification, RFC 7946.

Handling improperly formatted spatial data

When you load spatial data from other tools into BigQuery, you might encounter conversion errors due to invalid WKT or GeoJSON data. For example, an error such as Edge K has duplicate vertex with edge N indicates that the polygon has duplicate vertices (besides the first and last).

To avoid formatting issues, you can use a function that generates standards-compliant output. For example, when you export data from PostGIS, you can use the PostGIS ST_MakeValid function to standardize the output. Alternatively, import your data as text and then convert it by calling ST_GEOGFROMTEXT or ST_GEOGFROMGEOJSON with the make_valid parameter. When make_valid is TRUE, these functions attempt to repair invalid polygons.

To find or to ignore the improperly formatted data, use the SAFE function prefix to output the problematic data. For example, the following query uses the SAFE prefix to retrieve improperly formatted spatial data.

SELECT
  geojson AS bad_geojson
FROM
  mytable
WHERE
  geojson IS NOT NULL
  AND SAFE.ST_GeogFromGeoJson(geojson) IS NULL

Constraints

Geospatial analytics does not support the following features in geospatial formats:

  • Three-dimensional geometries. This includes the "Z" suffix in the WKT format, and the altitude coordinate in the GeoJSON format.
  • Linear reference systems. This includes the "M" suffix in WKT format.
  • WKT geometry objects other than geometry primitives or multipart geometries. In particular, geospatial analytics supports only Point, MultiPoint, LineString, MultiLineString, Polygon, MultiPolygon, and GeometryCollection.

See ST_GeogFromGeoJson and ST_GeogFromText for constraints specific to GeoJson and WKT input formats.

Load Google Earth Engine geospatial data

Google Earth Engine is a geospatial data platform that compiles and analyzes insights from satellite and earth observation imagery using raster data, where data is organized over a grid of cells representing information on digital images. While BigQuery primarily works with tabular vector data, users can use their BigQuery data in conjunction with raster data from Earth Engine to incorporate both vector and raster datasets in their workflows.

For information about exporting Earth Engine data to BigQuery, see Exporting to BigQuery.

Transforming geospatial data

If your table contains separate columns for longitude and latitude, you can transform the values into geographies by using GoogleSQL geography functions such as ST_GeogPoint. For example, if you have two DOUBLE columns for longitude and latitude, you can create a geography column with the following query:

SELECT
  *,
  ST_GeogPoint(longitude, latitude) AS g
FROM
  mytable

BigQuery can convert WKT and GeoJSON strings to geography types. If your data is in another format such as Shapefiles, use an external tool to convert the data to a supported input file format, such as a CSV file, with GEOGRAPHY columns encoded as WKT or GeoJSON strings.

Partitioning and clustering geospatial data

You can partition and cluster tables that contain GEOGRAPHY columns. You can use a GEOGRAPHY column as a clustering column, but you cannot use a GEOGRAPHY column as a partitioning column.

If you store GEOGRAPHY data in a table and your queries filter data by using a spatial predicate, ensure that the table is clustered by the GEOGRAPHY column. This typically improves query performance and might reduce cost. A spatial predicate calls a boolean geography function and has a GEOGRAPHY column as one of the arguments. The following sample shows a spatial predicate that uses the ST_DWithin function:

WHERE ST_DWithin(geo, ST_GeogPoint(longitude, latitude), 100)

Using JOINs with spatial data

Spatial JOINs are joins of two tables with a predicate geographic function in the WHERE clause. For example:

-- how many stations within 1 mile range of each zip code?
SELECT
    zip_code AS zip,
    ANY_VALUE(zip_code_geom) AS polygon,
    COUNT(*) AS bike_stations
FROM
    `bigquery-public-data.new_york.citibike_stations` AS bike_stations,
    `bigquery-public-data.geo_us_boundaries.zip_codes` AS zip_codes
WHERE ST_DWithin(
         zip_codes.zip_code_geom,
         ST_GeogPoint(bike_stations.longitude, bike_stations.latitude),
         1609.34)
GROUP BY zip
ORDER BY bike_stations DESC

Spatial joins perform better when your geography data is persisted. The example above creates the geography values in the query. It is more performant to store the geography values in a BigQuery table.

For example, the following query retrieves longitude, latitude pairs and converts them to geographic points. When you run this query, you specify a new destination table to store the query results:

SELECT
  *,
  ST_GeogPoint(pLongitude, pLatitude) AS p
FROM
  mytable

BigQuery implements optimized spatial JOINs for INNER JOIN and CROSS JOIN operators with the following GoogleSQL predicate functions:

Spatial joins are not optimized:

  • For LEFT, RIGHT or FULL OUTER joins
  • In cases involving ANTI joins
  • When the spatial predicate is negated

A JOIN that uses the ST_DWithin predicate is optimized only when the distance parameter is a constant expression.

Exporting spatial data

When you export spatial data from BigQuery, GEOGRAPHY column values are always formatted as WKT strings. To export data in GeoJSON format, use the ST_AsGeoJSON function.

If the tools you're using to analyze the exported data do not understand the GEOGRAPHY data type, you can convert the column values to strings using a geographic function such as ST_AsText or ST_AsGeoJSON. Geospatial analytics adds additional points to the line where necessary so that the converted sequence of edges remains within 10 meters of the original geodesic line.

For example, the following query uses ST_AsGeoJSON to convert GeoJSON values to strings.

SELECT
  ST_AsGeoJSON(ST_MakeLine(ST_GeogPoint(1,1), ST_GeogPoint(3,2)))

The resulting data would look like the following:

{ "type": "LineString", "coordinates": [ [1, 1], [1.99977145571783, 1.50022838764041], [2.49981908082299, 1.75018082434274], [3, 2] ] }

The GeoJSON line has two additional points. Geospatial analytics adds these points so that the GeoJSON line closely follows the same path on the ground as the original line.

What's next